Hinge Joint Structural Damage Identification Of Soft Yoke Mooring System Based On Deep Learning

As a positioning system of FPSO,Soft Yoke Mooring System(SYMS)is used in the south China sea and Bohai sea of China because of its advantages.Because it is subjected to the combined action of wind,wave and current for a long time,it is easy to cause fatigue damage of hinged joints.The hinge joint is an important structure in the SYMS.If hinge points are damaged,it will seriously affect the mooring capacity of FPSO and even cause safety problems.Therefore,it is very important to study damage identification of Soft Yoke Mooring System.Because the traditional numerical method has many shortcomings,the ability of this method to identify the damage state of hinged joints is not good.In this paper,an analytical method combining deep learning with data feature processing is proposed to research the damage identification of hinge points in SYMS.Firstly,this paper introduces the deep learning method,which is often used in classification problems,and the application process of deep learning method in damage identification of Marine structures is explained.The finite element simulation model of SYMS is constructed,and we need analyzing the motion behavior of SYMS,selecting the effective eigenvalues according to the analysis results.The damage identification of hinge joints is studied by using multi-layer feedforward neural network method.the verification results show that the depth method can effectively identify the damage state,and the highest verification accuracy is 81.94%.In order to verify the effectiveness of the method in real structure,building the large-scale experimental simulation system of SYMS.Marine natural and sampling conditions is used for the training,testing and verification of deep learning models.The deep learning method and feature processing method are combined to train and verify the damage recognition model.Finally,comparing the effects of different feature inputs and different deep learning architectures on the generalization ability by experimental results.The results show that feature selection and extraction can improve the generalization ability of the damage identification model.And comparing with the multi-layer feedforward neural network,the damage identification method based on deep stacked sparse autoencoder network can accurately identify the location and damage state of hinge nodes.recognition accuracy of the deep stacked sparse autoencoder network for the verification set is 84.67%,which proves the accuracy of the deep learning method in the identification of Marine structural damage.